Written by Ted Hoff, PhD and edited by Alan J. Weissberger
Please refer to earlier post on Errors and Corrections to Malone’s book.
Malone omits why I was hired at Intel as I was not a chip designer. The reason was Bob Noyce’s view that LSI circuits were moving in the direction of systems on a chip, and that Intel should have some in-house systems engineering expertise. Noyce asked Jim Angell, a EE Professor at Stanford who had consulted for Fairchild, to recommend some candidates. I believe he gave Noyce three names, and I was the only one of those not working for Fairchild. Noyce then called to invite me to join Intel.
# ] PAGE OBJECTION/SIGNIFICANCE
1] 54 Gordon Moore’s reflow patent covered a significant step in making the silicon gate process manufacturable by preventing cracking of the metal layer and thereby improving yield.
2] 55 Malone mentions Max Palevsky, but omits his connection to Scientific Data Systems, of which he was a co-founder in 1961. He sold the company to Xerox in 1969. His opinions were especially valuable, coming from the perspective of a computer company executive.
3] 121 Creative confrontation should be attributed to Andy Grove–and the emphasis should be on “creative.”
4] 128 Andy Grove played a major role in establishing that R&D had to work hand in hand with production. All three men said that at Fairchild, R&D would develop some new process only to have production resist adopting it.
5] 129 Andy Grove instituted “management by objectives,” which required progress reports to include predictions of future developments as well as past accomplishments. In subsequent reports, accomplishments had to be compared with the prior predictions. That discipline helped Intel’s engineers and management become much better at predicting progress and scheduling product development.
6] 134 Why does Malone say that the 3101 semiconductor memory came out in less than 18 months after Intel’s founding when it was approximately one year?
7] 136 Malone’s quote of Gordon Moore’s description of MOS is somewhat garbled. Gordon was talking about Intel’s MOS, which used the silicon gate process, not most other companies’ MOS products, which used metal gate.
8] 138 Malone mentions flash memory, but how was it developed?
9] 138 Malone describes Gordon Moore’s goal for an 1102 DRAM replacement to have no overlap with the Honeywell design, but makes no mention of how that was to be done. See page 156 below for more details.
10] 146 Malone omits the differences between a calculator chip set and a CPU oriented chip set. Consider the typical calculator set using a printer, and including a printer control chip. Should it be desired to use a different printer, a new printer control chip would be required–involving circuit design, a new chip layout, chip manufacturing and testing, as well as circuit board design to use the new printer controller chip. Those steps might take months, and involve tens of thousands of dollars of research and development. With a CPU approach, the new printer would typically require an afternoon of programming, the wiring of a new connection cable and the burning of a new EPROM–all taking less than a day.
11] 149 Malone omits why Intel would consider undertaking a custom chip job–which had the potential of delaying its development of semiconductor memory, Intel’s primary corporate goal. The reason Intel took the custom chip project (from Busicom) was concern about the rate at which semiconductor memory would be adopted by its target customers and to generate needed revenue in the interim. A custom job was expected to generate revenue much more quickly than semiconductor memory components.
12] 149 The details of the Busicom agreement–60,000 kits, price per kit not to exceed $50.00 should be presented at this point in the narration, i.e. at the time of the April, 1969 agreement, not later.
13] 151 In the early days of Intel, Bob Noyce frequently discussed many concepts–various aspects of computers, possible ways to implement bipolar ROM (one led to a patent), and other new and useful ways to use semiconductor technology. Those discussions had nothing to do with microprocessor chip set definition, design or development.
14] 153 Malone claims Bob Noyce had gone renegade–but omits what Bob should have done regarding the Busicom agreement and its problems. Should he have abandoned the agreement and walked away from a multi-million dollar order that Intel needed, or should he have just gone ahead and risked bankrupting the company trying to fulfill the Busicom agreement?
15] 153 Malone is incorrect in stating I felt I lacked skill in software. He omits that Intel’s MOS design team used a simulation program I had written in much of their design work. That program, initially run on an outside time-sharing service, was used so much that eventually Intel purchased a PDP-10 computer system primarily to run that simulation program. I have a copy of a memo I had written in 1975, discussing the merits of some proposed changes to the program. That proves the program was still being used seven years after Intel started operations.
16] 153 Malone omits the series of steps proposed in an effort to simplify the Busicom chip set. They ultimately led to the 4004 architecture (which was not a copy of a minicomputer as Malone claims). Those steps included breaking down the floating point arithmetic into digit by digit steps by making greater use of ROM; then noting that BCD arithmetic could be done by combining a binary step with BCD correction, thereby allowing the underlying processor to be a binary device (simpler than BCD); replacing shift-register memory with DRAM, which would allow simplifying the memory control logic while speeding up operations and reducing transistors per bit from six to theree; and noting that a simple binary processor with ROM could perform many of the operations being implemented by separate chips.
Also omitted is a definition of a CPU as used at Intel: a CPU has two major sections: one performs program sequence control with instruction fetching and interpretation; the second performs data manipulation as specified by each fetched and interpreted instruction. Bit slice chips only performed a portion of the latter function.
17] 153 Malone mentions that the Intel concept developed by Stan Mazor and myself would be a 4-chip design, but omits that it was an alternate to Busicom’s 10 to 12 chip design yet would still perform all the functions provided by the Busicom set.
18] 154 Malone describes Bob Graham’s letter to Busicom of September 16, 1969, as a “note.” He omits that it included price quotes for kits based on Busicom’s chip set and for Intel’s proposed chip set. It also included specifications for Intel’s chip set as well as the proposed instruction set for what would become the 4004 CPU chip.
19] 154 In saying that the Intel approach would involve a lot of sofware yet to be written, Malone omits that the Busicom chip set also needed extensive software. It was expected that the routines replacing floating point arithmetic and the I/O chips would be written once, coded into one or more ROMs, and then that standard ROM set would be used in the various calculator models.
20] 156 Malone mentions the 1102 1K DRAM problems and erroneously states the 1103 1K DRAM would be based on the 1102 core. Les Vadasz had told me of the difficulty caused by the 1102s need for an intermediate voltage generator. That generator was required because the 1102 used a single word line that required three different operating levels: unselected, read selected and write selected. Generating the read selected level was the problem, so I suggested to Les that by using 2 word lines, i.e. separate read and write, the intermediate level could be eliminated, but at the cost of a somewhat larger memory cell. Les liked the idea and urged me to file for a patent on that approach. The patent issued may have been Intel’s first. It soon became known that certain types of coupling on the 1103 DRAM chip could cause memory loss due to induced bipolar transistor action–solved by adding substrate bias and using an 18-lead package. I understand my patent was the only one disclosing a 3-transistor DRAM cell for which substrate bias could be applied.
21] 166 In discussing some events of 1970, Malone omits that in January, 1970 Bob Graham and I predicted that semiconductor memory would replace magnetic core menory in an Electronic Products magazine article. This article states that semiconductor memory would be priced below a penny per bit by 1972, at which point magnetic core memory could not be able to compete (despite its advantage of inherent non volatility).
22] 179 Malone mentions using established programming languages to make microprocessors work, but omits how that was accomplished. Intel found various ways to develop support for them–including Gary Kildall’s development of PL/M.
23] 179 In discussing marketing for the microprocessors, Malone cannot seem to understand that there were markets other than replacing mainframe computers or minicomputers. Within Intel, we were developing a story to tell to a new class of customers. Faggin and I had discovered that the MCS-4 chip set was fantastic for solving logic design problems, e.g. Faggin had to build testers, and I had to build programmers for PROMs and EPROMs. We concluded that, if we as engineers find microprocessors so useful, there will be many more engineers out there that would feel the same way.
24] 182 Bob Graham’s “Intel Delivers” campaign included a policy of not talking about a product until it was on distributor’s shelves. Until that happened, many LSI products from other companies were touted, but were not (or never became) available.
25] 187 Malone, in discussing the goal of increasing the clock speed for an upgrade to the 8008, fails to adequately cover the role of Intel’s newly developed n-channel silicon gate process–inherently faster than the previous p-channel technology and the primary reason that clock rates could be increased.
26] 188 Malone mentions Intel’s board having objections to the microprocessor as diverting focus from semiconductor memory product development. One reason for the board’s concern was that entering the computer business could be perceived by memory customers as competition for their business. Those of us who wanted the microprocessor announced argued that we could avoid that impression if we did not tout microprocessors as replacing our computer customer’s systems. It soon became evident that even big computer vendors needed little controllers throughout their systems. Many had such processors on their drawing boards at the time Intel came out with its microprocessors. Those computer customers quickly adopted Intel microprocessor devices to meet their needs. Consider that even the IBM PC used a microcontroller just to communicate with its keyboard. If one purchased IBMs dot matrix printer with an IBM PC, that printer also contained a microcontroller. Thus an IBM PC with that printer would have two embedded controllers compared to one PC type microprocessor.
27] 191 Malone argues that marketing produced endless manuals, etc. The error in that statement was addressed in the previous ithistory.org blog post on “errors and corrections.” However, Malone omits that the user guide for the 4004 was written by Stan and myself.
28] 192 Malone expresses amazement that the EPROM was non-volatile, but omits what was really unique about it. All ROMs were non-volatile, but the EPROM was “field programmable” and could be reused after erasure. MOS ROMs were only available from a semiconductor manufacturer and were “mask programmed.” Bipolar PROMs were only programmable once. Therefore, the EROM was much more flexible for the user as it could be reprogrammed in the field and didn’t have to be discarded if there were changes to be made or programming errors (bugs) detected. Please see next omission on pg 192.
29] 192 Malone fails to understand the benefit of the EPROM. Before it, MOS ROMs had to be ordered from the semiconductor manufacturer–a process which could take weeks. The customer had to send his firmware to the factory which would make a mask containing that firmware, then wafers would be processed using that mask. Those wafers would be sorted, separated into chips, packaged and then tested again. Intel charged $600 for the first three units, then subsequent orders had a 50 piece minimum quantity at $25.50 per unit. With the EPROM a customer could debug his code and reprogram on the spot, a process taking less than a day, rather than weeks. The C1702 quartz-lid EPROM sold for $81 in single unit quantity (prices as of September 1972).
30] 192 Before Intel offered the Intellec development tools, it offered SIM boards which had been developed by my group. They appear in the Sept. 1972 price list. Marketing had originally wanted to give them away, but I argued that we should sell them because customers would need such a device for their microprocessor development and it would cost them more to develop it themselves than what we could sell them for while still making a tidy profit. If we didn’t charge for development aids, we would ultimately see them as a burden on profits and discontinue them.
31] 192 Malone’s comment that Intel’s microprocessor development systems would have outsold personal computers neglects the cost difference. To compete in the PC market, Intel would have had to reduce prices by an order of magnitude from what they could charge for a development system, which was sold in small quantities (often just one) to each customer developing microprocessor applications.
32] 192 Intel’s 1972 annual report stated that its 1103 DRAM was the largest selling semiconductor memory in the world.
33] 193 From the time the 8008 microprocessor came out, its sales volume grew exponentially over time. We had expected those sales to drop away rapidly after the 8080 microprocessor was announced (and shipped to customers), but instead 8008 sales just stopped rising–they continued at a steady rate for quite some time.